I would assume it’s not directly scalable as there have been geofenced autonomous vehicles for years now (maybe not explicitly taxis, but I don’t think that matters too much). I think the real issue is that in the real world you just can’t hold as many variables constant, and we just don’t have the means yet to deal with that.

Its also possible that through this experiment they’ll learn something that helps them chip away at what things they need to hold constant.